Complaint: AI coding agents still produce poor UI even with references
A user reports trying multiple , including Codex, Claude, and Kimi, for UI work, and none produced something genuinely usable. Even when given a , the agents' s came out poorly. The project already uses TailwindCSS, yet the agents frequently default to plain CSS and HTML instead of following that library.
Overall, the agents' UI and UX output is described as worse than what a junior developer or intern would produce. The post ends by asking whether there is a fix.
Key points
- User tried Codex, Claude, and Kimi for UI tasks with unsatisfying results
- Providing a did not lead to a good
- Agents tend to default to plain CSS/HTML instead of the project's specified library (TailwindCSS)
- The author rates the agents' UI/UX output as worse than a junior developer's